mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-02 07:13:02 +00:00
d2321581ec
* fix(cohere): emit streamed tool calls from v2 tool-call-start and tool-call-delta chunks * fix(cohere): type _parse_tool_call_delta without growing basedpyright counts The reassigned Final local and the untyped chunk lookups added basedpyright errors past the base ceiling. Narrow them with casts; per-rule counts for the file are now at or below the merge-base. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * fix(cohere): parse streamed tool calls with a typed model and map Cohere TOOL_CALL to tool_calls Read the streamed tool call through a frozen pydantic model instead of casts, which the strict and type-discipline gates refuse. Map Cohere's TOOL_CALL and STOP_SEQUENCE finish reasons, so a stream that reads Cohere's message-end still ends tool_calls. The regression test now drives litellm.acompletion with two parallel streamed calls and checks the rebuilt calls and the final finish reason * fix(cohere): close a streamed tool call that sent no arguments with an empty JSON object --------- Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com> Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|---|---|---|
| .. | ||
| chat | ||
| embed | ||
| rerank | ||
| __init__.py | ||